Metagenomics Reveals the Influence of Land Use and Rain onthe Benthic Microbial Communities in a Tropical UrbanWaterway

Gourvendu Saxena,a,b,h Suparna Mitra,c,d Ezequiel M. Marzinelli,d,e Chao Xie,d Toh Jun Wei,d Peter D. Steinberg,d,e

Rohan B. H. Williams,b Staffan Kjelleberg,d Federico M. Lauro,d,g Sanjay Swarupa,b,f,h

aDepartment of Biological Sciences, National University of Singapore, Singapore, SingaporebSingapore Centre for Environmental Life Sciences Engineering (SCELSE), National University of Singapore,Singapore, Singapore

cLeeds Institute for Biomedical and Clinical Sciences, University of Leeds, Leeds, United KingdomdSingapore Centre for Environmental Life Sciences Engineering (SCELSE), Nanyang Technological University,Singapore, Singapore

eCentre for Marine Bio-Innovation, School of Biological, Earth and Environmental Sciences, The University ofNew South Wales, Sydney, Australia

fNUS Environmental and Research Institute, National University of Singapore, Singapore, SingaporegAsian School of the Environment, Nanyang Technological University, Singapore, SingaporehSynthetic Biology for Clinical and Technological Innovation, National University of Singapore, Singapore,Singapore

ABSTRACT Growing demands for potable water have led to extensive reliance onwaterways in tropical megacities. Attempts to manage these waterways in an envi-ronmentally sustainable way generally lack an understanding of microbial processesand how they are influenced by urban factors, such as land use and rain. Here, wedescribe the composition and functional potential of benthic microbial communitiesfrom an urban waterway network and analyze the effects of land use and rain per-turbations on these communities. With a sequence depth of 3 billion reads from 48samples, these metagenomes represent nearly full coverage of microbial communi-ties. The predominant taxa in these waterways were Nitrospira and Coleofasciculus,indicating the presence of nitrogen and carbon fixation in this system. Gene func-tions from carbohydrate, protein, and nucleic acid metabolism suggest the presenceof primary and secondary productivity in such nutrient-deficient systems. Compari-son of microbial communities by land use type and rain showed that while there aresignificant differences in microbial communities in land use, differences due to rainperturbations were rain event specific. The more diverse microbial communities inthe residential areas featured a higher abundance of reads assigned to genes relatedto community competition. However, the less diverse communities from industrialareas showed a higher abundance of reads assigned to specialized functions such asorganic remediation. Finally, our study demonstrates that microbially diverse popula-tions in well-managed waterways, where contaminant levels are within defined lim-its, are comparable to those in other relatively undisturbed freshwater systems.

IMPORTANCE Unravelling the microbial metagenomes of urban waterway sedi-ments suggest that well-managed urban waterways have the potential to supportdiverse sedimentary microbial communities, similar to those of undisturbed naturalfreshwaters. Despite the fact that these urban waterways are well managed, ourstudy shows that environmental pressures from land use and rain perturbations playa role in shaping the structure and functions of microbial communities in these wa-terways. We propose that although pulsed disturbances, such as rain perturbations,influence microbial communities, press disturbances, including land usage history,

Received 2 October 2017 Accepted 6 May2018 Published 5 June 2018

Citation Saxena G, Mitra S, Marzinelli EM, Xie C,Wei TJ, Steinberg PD, Williams RBH, KjellebergS, Lauro FM, Swarup S. 2018. Metagenomicsreveals the influence of land use and rainon the benthic microbial communities ina tropical urban waterway. mSystems3:e00136-17. https://doi.org/10.1128/mSystems.00136-17.

Editor Ruth D. Gates, Hawaii Institute ofMarine Biology

Copyright © 2018 Saxena et al. This is anopen-access article distributed under the termsof the Creative Commons Attribution 4.0International license.

Address correspondence to Federico M. Lauro,FLauro@ntu.edu.sg, or Sanjay Swarup,dbsss@nus.edu.sg.

G.S. and S.M. contributed equally to this work.

Rain and land use influence microbialcommunities in well-managed urbanwaterways.

RESEARCH ARTICLEApplied and Environmental Science

crossm

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 1

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

have a long-term and stronger influence on microbial communities. Our study foundthat the functions of microbial communities were less affected by environmentalfactors than the structure of microbial communities was, indicating that core micro-bial functions largely remain conserved in challenging environments.

KEYWORDS benthic microbial communities, community composition and functions,land use, rain perturbations, urban environment

Nearly two-thirds of global annual rainfall is at or near the equator (1). This zone isalso the most populated part of the world, with 79% of the world population

residing in tropical regions, mainly in urban settings (2). Metropolitan areas in thetropics, such as Singapore and Hong Kong, meet their growing demands for drinkingwater by engineering systems for capturing and redistributing storm water. Thesesystems require the construction of extensive networks of paved drains and largecanals (3, 4), the sustainable management of which is a challenging task, especiallyusing ecological approaches. Water managers and urban planners around the world areimplementing water-sensitive urban designs for sustainable management of waterwaysto maintain or enhance the self-cleaning potential of waterways. However, suchpractices lack understanding of microbial processes, underpinning the environmentalservices (5) provided by these ecosystems.

Sediment-associated microbial communities provide diverse ecological services andare the drivers of many biogeochemical processes in surface water systems (6, 7). Microbialcommunities are able to degrade a wide range of organic contaminants (8) where nutrientlevels are low and available carbon sources are complex organic compounds (8). The abilityto degrade complex organic contaminants for energy harvest is important for the survivalof microbial communities in resource-deficient engineered urban waterways. These water-ways are typically limited in nutrients and simple sugars but rich in complex organiccompounds (9). As we move toward utilizing ecology-based approaches for sustainablewater resource management, characterizing the composition and functions of benthicmicrobial communities and how they respond to urban factors, such as land use types andrainfall, becomes important.

Urban storm water in engineered waterways passes through many different landuse types, such as residential and industrial areas. These areas contribute different loadsof sediments, chemicals, and microbial communities, which together affect the eco-system services of the engineered urban environment (10, 11) and downstream watercollection reservoirs. The rain events lead to the removal of settled sedimentarymicrobial communities by resuspension and immigration of fresh populations ofsoil-associated microorganisms into the waterways. Therefore, factors such as land usetypes and rain could have major influences on the assembly and functioning ofmicrobial communities in urban waterway systems. To the best of our knowledge, suchinfluences have not been characterized for urban environments using deep sequencingmetagenomic data.

In this study, we have examined a catchment in Singapore, a highly urbanizedtropical city with two major land use types, residential and industrial, to identify theinfluence of land use and rain on the benthic microbial communities in urban water-ways. These land use types covered �80% of the catchment (Fig. 1). We had previouslyfound the microbial communities to be similar within the same land use type, despitebeing present in spatially nonconnected regions (9). While site I4 receives storm waterand sediments exclusively from the eastern side of the waterway, which is an industrialland use type, site R10 receives storm water and sediments from residential land usetype on both sides of the waterway (Fig. 1). Water and sediments at these sites do notcross-influence each other. We therefore, selected these sites for analyzing sedimentarymicrobiomes in urban waterways. As rain is one of the major perturbation factors in thetropics, we further sampled two rain events with similar intensity but of differentdurations and different dry periods preceding the event.

Here, we first describe the near complete coverage of species composition of the

Saxena et al.

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 2

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

microbial communities of the catchment urban waterway system. To identify theinfluence of rain and land use on the microbial communities, we then compare thecommunity composition and functions, as deduced from the metagenomic data forsediments of microbial communities, sediments before and after rain events, andsediments of the two land use types. We compare the urban waterway microbialcommunities identified from our study with those of other freshwater systems to assessthe impact of low versus high contaminant levels on the diversity of microbial com-munities in urban waterways.

RESULTSUrban catchment. The catchment in this study covers an area of about 23 km2 and

has two major land use types, residential and industrial (Fig. 1). The regions ofresidential land use type are more dispersed in the catchment, compared to theindustrial areas, which has only one region (Fig. 1). The extensive subnetwork on drainsand canals in this catchment consists of more than 16,000 m of big and intermediatecanals (�20 m wide) and a dense network of small drains of length more than 300 km,which collect storm water from the local watershed and drain into Pandan reservoir.Therefore, the sedimentary microbial communities in the waterways reflect the micro-bial communities in the catchment. We collected sediment samples from six differentlocations at each land use site (Fig. 1) before and after rain across two rain events (seeFig. S1A in the supplemental material). This resulted in a total of 48 sediment samples.

Environmental pressures and rain intensity. In our previous work, we character-ized 48 environmental parameters from the same catchment (9). Using this data, weestablished correlations between the rain intensity and environmental parametermatrix. The results revealed that dissolved oxygen and temperature have a positivecorrelation with the total suspended solids, while sodium and potassium display anegative correlation with the monthly rain intensity (P � 0.05) (Fig. 2Ai). Metals, suchas copper, cobalt, and zinc, and organic contaminants, including styrene, were signif-icantly higher in industrial areas than in residential areas (Fig. 2Aii).

In 2011 and 2012, the catchment received annual rainfalls of 3,423 mm and2,700 mm with 190 and 170 rainy days and an average rainfall of 18 and 16 mm per rain

N

FIG 1 Model watershed and sampling locations. The inset shows Pandan watershed in the southwesternregion of Singapore. The sampling locations were selected from representative and physically isolatedresidential and industrial land use types. (Adapted from reference 9 with permission of the publisher.)

Urban Waterway Microbial Communities

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 3

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

event day, respectively (Fig. 2B). The rain data from October 2011 to April 2012 areshown in Fig. 2B. Our sampling in January falls in the northeast monsoon season(December-March), with relatively frequent dry-wet fluctuations. The rain intensities ofrain event 1 (RE1) (15 January 2012) and rain event 2 (RE2) (20 January 2012) were both16.25 mm. However, RE1 was of longer duration (4 h) than RE2 (2 h). The dry periodprior to RE1 was about 20 days (more than 10-mm rain), while it was 5 days for RE2(Fig. 2B).

Microbial community composition in the urban waterway environment. Se-quencing of metagenomic DNA fragments using the Illumina HiSeq2000 platformresulted in 1,554,726,091 paired reads with a read length of 101 bp for the paired reador a total of 3,109,452,182 raw sequences (a total of 314,054,670,382 bp). Qualitytrimming to remove low-quality reads resulted in 3,023,688,898 (97.2%) reads (Fig. S1B).A total of 1,460,323,652 (48.3%) reads, with an average of 30,423,409 reads persample (standard deviation, 91,74,804) were successfully mapped against the Na-tional Center for Biotechnology Information (NCBI) nonredundant (nr) proteindatabase (12) (Fig. S1B). Rarefaction analysis of the 48 samples suggested thatsequencing depth achieved a high level of saturation of the taxonomic richness, as all

1 2

Rain

inte

nsity

(mm

)

16.5 mm

01/1007/1013/1019/1025/1031/1006/1112/1118/1124/1130/1106/1212/1218/1224/1230/1205/0111/0115/012 0/0126/0101/0207/0213/0219/0225/0202/0308/031 4/0320/0326/0301/0407/0413/0419/0425/04

Rain

inte

nsity

(mm

)

Rain event 2

12:0

013

:00

14:0

015

:00

16:0

017

:00

18:0

0

0

2

4

6

8

10

12

14

16

01:0

002

:00

03:0

004

:00

05:0

006

:00

07:0

0

Rain event 1Time (hours)

T = 4hrs

T = 2hrs

ΔT = 2hrs

0

140

20

40

60

80

100

120

2011 2012

30/04

Ai

B

0.10.40.71

ResidentialIndustrial

JanFebMarApr

MayJunJul

AugSepOctNovDec

300

150

0

Diss

olve

d ox

ygen

Tem

pera

ture

Sulfa

tes

Sodi

um

Pota

ssiu

mM

onth

ly

rain

fall

(mm

)

AiiCorr 0.85 0.48 -0.45 -0.42 -0.45

Conc

entr

a�on

(ppb

)0

5

10

15

20

25

Copp

erCo

balt

Zinc

Styr

ene

Tota

l org

anic

carb

on Sodi

umPo

tass

ium

Nitr

iteCh

lora

mph

enico

l

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Conc

entr

a�on

(ppb

)

FIG 2 Environmental characteristics of the catchment. (Ai) Temporal variations of environmentalvariables, which showed significant correlations (Kendall tau, P � 0.05) with rain. (Aii) Significantlydifferent (P � 0.05) environmental variables between the two land use types are shown. Error bars depictstandard errors. (B) Distribution of daily rain events from October 2011 to April 2012. The inset graphshows the hourly rain intensity data for the two rain events.

Saxena et al.

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 4

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

the plots reached asymptote, with the exception of one sample (Fig. S1C). Details of theread statistics for sequencing at each data processing stage are provided in Table S1.

The combined metagenome of benthic microbial communities in urban waterwayswas composed of 75 phyla and 4,163 species, based on reads mapped to NCBI nrdatabase. There were 1,116 genera identified that could be discriminated with a higherlevel of confidence. More than 99% of the mapped reads in the sediments of urbanwaterways were represented by 23 phyla (~30%) (Fig. 3Ai). Almost 80% of the anno-tated metagenome reads belonged to the top four phyla: Proteobacteria, Cyanobacteria,Bacteroidetes, and Actinobacteria. Hierarchical clustering showed different trends forthe taxonomic groups. Four of these clusters (Fig. S2) can be explained by factors suchas land use or rain events. Table S2A lists the top 10 genera in these clusters. Clusters1, 3, and 4 represent taxonomic groups that were enriched in either land use type. Thisgroup showed high abundance of genera involved in N-S cycling, such as Nitrosospira,Nitrosovibrio, Nitrosopumilus, Beggiatoa, and Sulfuritalea. Cluster 2 represents taxonomicgroups that appeared only after the rain event. These include reads from viruses withdiverse hosts such as Asfivirus and Phaeovirus, the oral cavity commensal Alloprevotella,and the deep ocean genus Zunongwangia.

Functional potential of microbial communities. Microbial communities of urbanwaterways had a high abundance of genes related to carbon metabolism, which wassignificantly higher in residential areas (P � 0.05) than in the industrial areas (Fig. S3A),based on t test with standard Bonferroni corrected P values. The second highestabundant genes were genes related to virulence factors and protein and nucleic acidmetabolism (Fig. S3A). There was no significant difference between the two land usetypes for other gene categories, except metabolism of aromatic compounds (highabundance in residential areas; P � 0.05), membrane transport (high abundance inresidential areas; P � 0.01), sulfur metabolism (high abundance in residential areas; P �

0.01), and photosynthesis (high abundance in industrial areas; P � 0.01) (Fig. S3A).Different trends of functional gene abundance were evident from the hierarchicalclustering results. Three of these clusters (Fig. S3B) are explained by factors such as landuse or rain events. Table S2B lists the top 10 functional genes in these clusters, basedon abundance. Cluster 1 largely represents functional genes enriched in the microbialcommunities from RE1. The genes are involved in biosynthesis, e.g., biosyntheses offlavonoids and carotenoids, organic remediation, such as xylene degradation, andsecretion, such as bacterial secretion system. Cluster 2 showed functions enrichedmainly in industrial samples, which included functions related to the infection processof pathogens.

Influence of land use on microbial communities. The microbial communities fromthe two land use types were clearly separated when analyzed via neighbor-net net-works from metagenomic DNA sequences and MEGAN taxonomic annotation (Fig. 3B).To further test the network robustness, a read-level bootstrapping of network cluster-ing was conducted with 100 iterations from random subsamples of the data sets asdescribed in Materials and Methods. The residential and industrial samples were clearlyand reproducibly separated, with the residential group showing a greater within-groupvariation relative to the industrial samples (Fig. 3B).

At the genus level, most reads were from Nitrospirae (Nitrospira) and Cyanobacteria(Coleofasciculus) (Fig. 3Aii). The reads for these taxonomic groups, which are involvedin nitrogen cycling and carbon fixation, were more abundant in industrial areas than inresidential areas (P � 0.05 by t test) (Fig. 3Aii). Reads from genera with low abundancein the catchment, such as “Candidatus Solibacter,” Anaeromyxobacter, Haliscomenobac-ter, Geobacter, and Anaerolinea represented a higher relative proportion in the com-munity from residential areas (Fig. 3Aii). The distribution of taxonomic groups (genera)in different samples suggested that the benthic microbial communities in residentialland use were more diverse and dispersed than in industrial sediments (P � 0.05 by ttest) (Fig. 3C). The distribution of taxonomic groups (genera) provided further evidencethat the benthic microbial communities in residential land use were more diverse, as

Urban Waterway Microbial Communities

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 5

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

the number of unique taxa was greater in residential sediments with approximatelytwice as many genera in residential sediments (81 genera) than in industrial sediments(42 genera) (Fig. 3C).

These results were further substantiated by permutational multivariate analysis ofvariance (PERMANOVA) analyses (Table S3A). The influence of land use was moredominant than rain perturbation on the structure of microbial communities. Pairwiseresults for 16S rRNA-based taxonomic profiles showed communities from both land usetypes were significantly different in both rain events (Fig. 3B; see Table S3A also).Permutational multivariate analyses of DISPersion (PERMDISP) analysis identified sig-nificant dispersion between the microbial communities from residential and industrialland use types (Table S4).

The distribution of functional gene abundances also showed clear separation be-tween the two land use types (Fig. 4A). The number of unique functions was greaterin residential sediments, with approximately four times as many functional genes (467genes) compared to industrial sediments (161 genes) (Fig. 4B). The most abundant andsignificantly different functional genes between the two land use types were engagedin organic remediation, such as 1,2-dihydroxynaphthalene dioxygenase and energyprocesses, including heme transporter IsdDEF and ferredoxin oxidoreductase PebA inindustrial areas, mostly from cyanobacteria. In residential areas, of the significantlydifferent functional genes, the most abundant were related to competition, such aspolymyxin resistance (ArnA) and energy processes, including hydrogenase-4 compo-nent F, heterodisulfide reductase, and NAD-independent deacetylase AcuC (Fig. 4B).

These results were further substantiated by PERMANOVA analyses (Table S3B),where the influence of land use was more dominant than rain perturbation on thestructure of microbial communities.

C

B

0

0.25

0.50

0.75

1.00

1.25

ResidentialIndustrial

Num

ber

of r

eads

(x1

07 )

*

** * *

** *

I4Pr1I5Pr1

I5Po1I6Po1

I4Po1R6Po1

R5Po1

R4Po1I2Po1

I3Po1

I1Po1

I3Pr1

I1Pr1I2Pr1

I3 Po2

I1Po2I2Po2

I2Pr2

I1Pr2I3Pr2

I4Po2I5Po2I6Po2

I5Pr2I6Pr2

I4Pr2

R6Pr2R5Pr2

R4Pr2R6Po2

R5Po2

R4Po2R2Po1

R1Po1

R3Po1

R2Pr1

R1Pr1R3Pr1

R5Pr1

R6Pr1

R4Pr1 R3Pr2

R1Pr2R2Pr2

R2Po2R3Po2

R1Po2

I6Pr1

0.01

Residential RE2Residential RE1

Industrial RE2Industrial RE1 Pre-rain

Post-rain

Ai

Land

-use

(LU

)

Rain-event (RE)

RE

2

RE

2R

E1

N

0

30

70 R−RI−II−R

Aii

Residential Industrial

FIG 3 Structure of benthic microbial communities of residential and industrial land use types before and after two rain events in urban waterways are shown.(Ai) Relative abundances (as percentages) of benthic microbial communities at the phylum level in waterways. (Aii) Average abundance of top 20 bacterialgenera, sorted based on abundance in the two land use types. Values that are significantly different (P � 0.05) in abundance in the two land use types aredenoted by an asterisk. (B) The neighbor-net network of samples from the two land use types (industrial [red] and residential [green]) based on Bray-Curtisdistance calculated using microbial community profiles at the species level. (Inset) Summary of distances within and between groups of microbiome profilesat the species level of samples from the two land use types. (C) Taxon distribution between the two land use types and pre- versus postrain for the two rainevents. Lighter shades show differential genera (t test; P � 0.05, standard Bonferroni correction, permutations � 999) and darker shades show commonlyoccurring genera. Diversity indices are shown as follows: S, Shannon index (entropy); E, Buzas and Gibson’s evenness (*, P � 0.05).

Saxena et al.

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 6

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

Influence of rain on microbial communities. RE1 mediated a significant differencein the pre- versus postrain microbial community diversity in residential land usesediments (Fig. 3C). Moreover, the diversity of microbial communities increased signif-icantly (P � 0.05 by t test) from post-RE1 to the pre-RE2 in both land use types (Fig. 3C).The prerain sediment microbial communities had more unique genera than thepostrain sediment microbial communities, with slightly reduced numbers in RE2(Fig. 3C). The functions of the microbial communities differed significantly beforeand after rain only during RE1 (4-h duration), compared to RE2 (2-h duration) (Fig. 1)(PERMANOVA; Table S3). There was no difference in dispersion between pre- andpostrain microbial functions (PERMDISP; Table S4).

Functions of microbial communities within both land use types showed inversetrends before and after rain perturbations. In residential benthic microbial communi-ties, the unique functions decreased after rain events, while in industrial areas, theirnumbers increased after the rain events (Fig. 4B). The profiles of differential functions(before and after rain) remained similar for the two rain events in the benthic microbialcommunities of residential land use types. The majority of the most abundant func-tional genes (three out of five) in residential microbial communities belonged to

Residential RE2Residential RE1

Industrial RE2Industrial RE1

Pre-rainPost-rain

A

I2Po1

I3Po1I1Po1

R4Po1R5Po1

R6Po1

R2Po1

I4Pr2I3Pr1

R6Pr1

R5Pr1

R4Pr1

R1Po2R3Po2R2Po2

R2Pr2

R1Pr2

R3Pr2

I5Pr1

I4Pr1I5Po1

I6Po1

I4Po1I6Pr1

I6Pr2I5Pr2I6Po2I4Po2

I5Po2I2Pr1

I1Pr1I3Pr2

I1Pr2

I2Pr2 I2Po2I1Po2I3Po2

R1Pr1R3Pr1

R2Pr1

R4Po1R3Po1R6Po2

R5Po2R6Pr2R5Pr2

R4Pr2

0.01

R4Po2

B

FIG 4 Distribution of microbial functions in urban waterways. (A) The neighbor-net network cluster of samples from the two land use types based on Bray-Curtisdistance calculated using microbial community functional potential annotated through SEED databases. (B) Functional gene distribution between the two landuse types and pre- versus postrain for the two rain events. Lighter shades show differential genera (t test; P � 0.05, standard Bonferroni correction, permutations �999) and darker shades show commonly occurring genera. Functional gene categories are defined by most abundant functional genes. The functional genes inrespective categories are annotated with numbers. Taxonomy assignments of genes in different groups are shown as lowercase letters.

Urban Waterway Microbial Communities

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 7

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

functions associated with competition, such as outer membrane porins, polymyxin andmultidrug resistance prerain, and were replaced by genes belonging to biomassgrowth, such as tRNA-ribosyl transferase and methenyltetrahydrofolate cyclohydrolasein postrain microbial communities (Fig. 4B). Despite the similarities in the functionalpotential across the two events, they were derived from different sets of taxonomicgroups (Fig. 4B). In the industrial benthic microbial communities, the most abundantfunction of those that differed before and after rain changed from organic remedia-tion, such as ethylbenzene dehydrogenase, biomass growth, including phosphoetha-nolamine transferase and nitrate reductase to �-carotene hydroxylase and phosphatestarvation-inducible ATPase during RE1 (Fig. 4B). During RE2, the majority of the mostabundant functional genes before rain were related to primary production, such asheme transporter IsdDEF and chlorophyll a(b) binding. These genes were replaced byviral marker genes and biofilm-related genes, such as prophage clp protease, majorcapsid protein, and the type IV pilus (PilE) (Fig. 4B). As for the residential areas, thefunctional genes identified in the two rain events were attributed to different sets oftaxonomic groups. Functions for genes from certain genera, such as Thauera andNovosphingobium, were related to organic remediation, while Clostridium containedgenes involved in carbon and nitrogen fixation distributed between land use types andthe two rain events (Fig. 4B).

Benthic community structure is similar to undisturbed freshwater systems. Wethen compared the microbial communities from urban waterways with that of a naturalfreshwater system (river) published elsewhere (13). The samples in that study werecollected from a pristine site of the river (upstream of urban influence) and a siteinfluenced by industrial effluents on the same river (downstream of urban influence)(13). Diversity analysis of upstream and downstream sedimentary microbial communi-ties (13) revealed that benthic microbial community profiles decreased in diversityvalues with increased exposure to urban influences. The pristine site (13) had the mostdiverse microbial communities, followed by the microbial communities of both landuse types in urban waterways of the present study. The microbial communities from thesite influenced by industrial effluents had the least diversity compared to the pristinesite and urban waterways (P � 0.001 by Mann-Whitney U test; n � 3) (Fig. 5A). Theabundance of primary producers, such as cyanobacteria, which is highest at the pristinesite of the river (37%), was maintained in the waterway system (10%) of our study.However, primary producers were completely outnumbered in the downstream system(Fig. 5A). The differences in the relative abundance of the top five phyla in the naturalfreshwater system and our study show that genera involved in primary productivity(Cyanobacteria) and carbon cycling (Actinobacteria) are reduced in microbial commu-nities from urban influenced sites, whereas heterotrophs (including Proteobacteria,Bacteroidetes, and Verrucomicrobia) increase in abundance (Fig. 5A). Within the Proteo-bacteria, the abundance of Alphaproteobacteria was reduced and the abundance ofBetaproteobacteria increased in microbial communities under urban influence (Fig. 5B).However, the benthic microbial communities in the urban waterway system maintainedthe structure of Alpha- and Betaproteobacteria close to the benthic microbial commu-nities of undisturbed natural freshwater (Fig. 5C).

DISCUSSION

To understand the influence of urban factors, such as land use, and rain on thebenthic microbial communities in a tropical urban waterway, we focused on an~25-km2 catchment area that encompasses two major land use types before and afterrain events. As sediments from eroding soil enter the waterways after rain events, thesedimentary microbial communities are exposed to multiple stress factors. Thesefactors include land use-specific chemical stressors (14–16) and different shear forcesresulting from increased flow after rain (17–19), which are dependent on both sedimentand rain properties. The current experimental design allowed us to study the influenceof the above two factors, namely, land use types and rain on the benthic microbialcommunities. By adopting next-generation sequencing (NGS), we sequenced the met-

Saxena et al.

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 8

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

agenome to near complete coverage of species, and detected and identified nearlyall microbial species, annotated in the NCBI nonredundant database, of the benthicmicrobial communities in this urban waterway (see Fig. S1 for nearly saturated rarefac-tion curve for all samples). The microbial communities in the urban waterways are verycomplex (rarefaction curve saturation [Fig. S1C] with �50% hits in the database[Fig. S1B]), and we have identified almost all the species possible using the currentdatabases.

The benthic microbial communities contained 75 different phyla, of which 23dominated the communities. We identified the taxonomic groups responsible for small,yet significant differences in the microbial communities between pre- and post-rainevents for the two land use types (Fig. 3C). The presence of Alphaproteobacteria,Cyanobacteria, and Actinobacteria in freshwater systems of a highly urbanized citylike Singapore is significant, as these taxa are highly sensitive to urban pressures,such as organic compounds and antibiotics (13, 20). Members of the microbialcommunities in the residential areas were more diverse (Shannon’s index; P � 0.05)and even (Buzas and Gibson’s evenness; P � 0.05) in microbial community structurethan those in industrial areas (Fig. 3C). In particular, a higher percentage ofColeofasciculus (Cyanobacteria)- and cyanobacterium-associated primary productivitygenes were observed in the industrial samples. This possibly results from the lowerconcentrations of antibiotics, such as chloramphenicol, observed in the industrial areas(9), as most cyanobacteria have been shown to be highly sensitive to antibiotics (20).The larger amounts of organic contaminants, such as styrene, could be stimulating thegrowth of xenobiotic-degrading strains of Anaeromyxobacter, Geobacter, and An-aerolinea in industrial areas (21, 22). Therefore, there is further scope for reducing theconcentrations of antibiotics and organic contaminants in residential and industrial

FIG 5 Taxon distribution in sediment microbial communities of natural river and urban waterway systems. (A) Comparison of phylum distribution of the twobenthic microbial communities in river water system (13) with urban water system at the phylum level. Diversity indices are shown as follows: S, Shannon index(entropy); E, Buzas and Gibson’s evenness (Mann-Whitney U test, n � 3, P � 0.001). (B) Trends of different classes in the Proteobacteria phylum between thethree microbial communities. (C) Distribution of different orders in the Alpha- and Betaproteobacteria classes of the Proteobacteria phylum in the three microbialcommunities.

Urban Waterway Microbial Communities

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 9

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

areas. Conditions favoring nitrite oxidation, such as the higher availability of nitritecoupled to a lower concentration of nitrate (9), might explain the proliferation ofNitrospira species in industrial areas. Certain Nitrospira species can utilize the completenitrate reduction to ammonium via nitrite (23), which may account for the higherabundance of Nitrospira in the urban waterways of this study.

Increase in the microbial diversity after rain event 1 (RE1) and before rain event 2(RE2) can be attributed to the higher availability of dissolved oxygen in water duringthe wet period. Increased temperature during the wet period, as shown by the positivecorrelation of water temperature with rain, might be due to the heated island effect(24). Storm water appears to mitigate the heated island effect but at the same timeexposes the microbial communities in waterways to elevated temperature. Therefore,it should be taken into consideration while planning to dissipate land surface heatusing rain water. A high abundance of heterotrophy- and biomass growth-relatedgenes in the microbial communities after rain reflects increased dissolved oxygen dueto frequent rain events in the catchment, while the presence of organic remediationgenes before rain appears to be linked to low levels of dissolved oxygen. Hence,microbial communities are responding to small changes in environmental conditionsthrough specific functional pathways, which can be exploited to implement specificmanagement interventions at the source to enhance the self water-cleaning potentialwaterways. Overall, the study found a clear separation between the metagenomicprofiles of the microbial communities belonging to residential and industrial sedimenttypes, both taxonomically and functionally. These results are consistent with studies ofmicrobial community structure and functions in comparable systems, such as oil-polluted environments (25), river systems with inputs from local land uses (16, 26), andnatural vegetation environments (27). The microbial communities in such environ-ments showed adaptation to the local environmental conditions both at the taxonomicand functional levels.

As this is the first comprehensive urban benthic microbial community metagenomicstudy, a comparison with information derived from other freshwater systems mayaddress whether urbanization exerts selective pressures that reduce microbial commu-nity diversity. Such a relationship was observed in Nanxijiang River near Wenzhou cityin the Zhejiang Province of China (13). The microbial communities of the urban watersystem studied here had Shannon diversity and Buzas and Gibson’s evenness indicessimilar to those from undisturbed Nanxijiang River communities, upstream of anurbanized area. The diversity indices of the urban waterway microbial communitieswere significantly higher than that of microbial communities from industrially influ-enced location of the same river. Such patterns suggests that well-managed waterways,that keep the contaminants levels below the permitted limits, can support microbialcommunities with a diversity and evenness comparable to those of undisturbed naturalfreshwater systems. Such diverse microbial communities can provide a wide range ofecological services and enhance the self water-cleaning capacity of waterways.

Reads mapped to diverse categories of amino acid and carbohydrate metabolismgenes were highly abundant in the microbial communities of the waterways, support-ing the notion of a large gene pool for utilizing a wide range of carbon sources. Theexistence of a flexible gene pool and its underlying phylogenetic structure is alsosupported by the high multivariate dispersion in the taxonomic analysis. The higherabundance of functional genes related to competition prior to rain events suggeststhat competition is driving the higher microbial community diversity in residential areas(Fig. 4B). The majority of microbial functions in the prerain residential sedimentarymicrobial communities have high abundance of, for example, outer membrane porinOprD, polymyxin resistance protein and multidrug resistance systems, which can beclassified as genes providing a competitive advantage. For example, OprD providescompetitive advantage to taxon groups through efficient uptake of nutrients (28).Likewise, polymyxin resistance protein (29) and multidrug resistance systems (30)provide a competitive advantage to taxonomic groups by either neutralizing antibioticsor by metabolizing the drugs as nutrients in such nutrient-deficient systems. In contrast,

Saxena et al.

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 10

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

the functions of microbial communities in industrial areas are more influenced byenvironmental pressures (anthropogenic chemicals), rather than within-communitycompetition, as suggested by the high abundance of ethylbenzene dehydrogenasegene (a common hydrocarbon-oxidizing enzyme). In summary, in urban waterways, theless diverse microbial communities might result from external environmental selectionpressures, while the more diverse microbial communities are influenced by intracom-munity competition.

Significant correlation between the taxonomic and functional profiles suggests thatthe functional potential may not be entirely independent of the structure of themicrobial communities (Table S5). The observed shifts in the functional potential inresponse to perturbations, therefore, appear to be a result of changes in the relativeabundances of taxonomic groups with some degree of functional redundancy ratherthan selection of more favorable adaptive traits.

With increasing urbanization and changing climate, this study provides insights intothe ability of microbial communities to maintain diversity in a manner comparable tothose of the natural freshwater systems, in response to pressures associated with landuse, rain, and management practices. Further, our study will also advance the devel-opment of management practices that benefit from ecologically based approaches toimprove surface water quality. Finally, it provides a microbial metagenomic data setfrom urban waterway sediments that can serve as a reference for microbial ecologystudies pertaining to urban environments.

MATERIALS AND METHODSSampling design. This study was designed to capture the taxonomic and functional diversity of

microbial communities in urban waterways and the influence of perturbation by rainfall events and landuse pressures on these communities. This study was conducted in the Ulu-Pandan catchment in thesouthwestern region of Singapore and drains water into the Pandan reservoir (Fig. 1). Sampling locationsrepresent a subset of all industrial and residential locations sampled in a previous study that usedmicroarray-based technologies and encompassed a wide range of environmental variables associatedwith changes in microbial communities (9).

Sediment samples were collected from the top 1-cm layer of sediments from the drains at anindustrial (I) location (1.3356° N, 103.7491° E) and a residential (R) location (1.3188° N, 103.7708° E; Fig. 1)before and after rain during two independent rain events (Fig. 2B). Samples were collected about 1 to2 h before rain (pre) and 3 to 4 h after rain (post), with a time lag of about 30 min between samplingat the two locations (Fig. 1). We followed the 2-h weather forecast provided by the National EnvironmentAgency of Singapore (http://www.nea.gov.sg/weather-climate/forecasts/2-hour-nowcast) to guide oursampling. Samples were collected during the two rain events: 15 January 2012 (rain event 1 [RE1]) and20 January 2012 (rain event 2 [RE2]) (Fig. 2B).

For each combination of timing (before rain versus after rain), rain event, and location, six indepen-dent sediment samples were collected (total of 48 samples; Fig. 2A) from the side drain of the canal,which collects the storm water from the local industrial or residential areas. It is worth noting thatalthough the industrial site sampled in this study is close to a residential area on the west, the water andsediment that it receives comes from the highly industrial eastern area (Fig. 1). Each sample was preparedby pooling sediment samples from five random sites within a 5-m diameter and collected in cleannuclease-free Falcon 50-ml conical centrifuge tubes (Thermo Fisher Scientific) using an autoclavedspatula. The 50-ml tubes were fully filled with the sediments, with about 5 ml of pore-water added. Thesamples were immediately transferred to an icebox and transported to the laboratory for DNA extraction.The time elapsed from the collection of the first sample to the beginning of DNA extraction was �2.5 h.

Rain data. The National University of Singapore (NUS) weather station, which is maintained by theDepartment of Geography, collects real-time meteorological data at 1-h intervals. Hourly rain intensitydata for the years 2011 and 2012 were processed to obtain total rain intensity per day.

DNA extraction and cleanup. Sediment samples were thoroughly mixed using a clean spatula priorto weighing. Sediments mixed with water were siphoned into 2-ml tubes and centrifuged to discard thepore-water. Sediments were then weighed to 200 � 20 mg for extracting DNA. DNA extraction wasperformed using the FastDNA spin kit for soil (MP Biomedicals, USA) according to the manufacturer’sprotocol. The purified DNA was eluted in nuclease-free water and cleaned using OneStep PCR inhibitorremoval kit (Zymo Research Corporation, USA) following the manufacturer’s protocol. DNA quantity wasquantified using a NanoDrop spectrophotometer (Thermo Scientific) and Quant-iT PicoGreen double-stranded DNA (dsDNA) assay kit (Invitrogen) according to the manufacturer’s protocol. Samples werestored in �80°C until further processing.

Illumina library preparation and sequencing for sediment samples. Prior to library preparation,the quality of the DNA samples was assessed on a Bioanalyzer 2100, using a DNA 12000 chip (Agilent).Sample quantitation was performed using Quant-iT PicoGreen dsDNA assay kit (Invitrogen). Next-generation sequencing library preparation was performed following Illumina’s TruSeq DNA samplepreparation protocol with the following modifications: to reduce the concentration of substances in the

Urban Waterway Microbial Communities

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 11

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

extracted DNA samples that may inhibit library preparation reactions, each DNA sample was purified withthe OneStep PCR inhibitor removal kit (Zymo Research Corporation, USA) prior to library preparation. Thepurified DNA samples were then sheared on a Covaris S220 system to ~500 bp (samples 31 to 44) and~300 bp (samples 291 to 324) (see Table S1 in the supplemental material), following the manufacturer’srecommendation. Size selection was performed on a Sage Science Pippin Prep instrument, using a 2%ethidium bromide agarose cassette and selecting for a tight spectral peak around 600 bp and 400 bp forsamples 31 to 44 and samples 291 to 324, respectively. Each library was tagged with one of Illumina’sTruSeq LT DNA barcodes to allow for library pooling prior to sequencing.

Invitrogen’s Picogreen assay was used for library quantitation, and the average library size wasdetermined by running the libraries on a Bioanalyzer DNA 7500 chip (Agilent). Library concentrationswere normalized to 4 nM and validated by quantitative PCR (qPCR) on a StepOne Plus real-timethermocycler (Applied Biosystems), using qPCR primers recommended in Illumina’s qPCR protocol, andIllumina’s PhiX control library as a standard. Libraries were then combined into four pools, and each poolwas sequenced in two lanes of an Illumina HiSeq2000 sequencing run at a read length of 101 bp for eachpaired end.

Data processing and analyses. Raw sequence reads for all 48 samples were quality trimmed/adapter clipped using cutadapt-1.2.1 (31) with default parameters (except for -m 30 -q 20). Thequality-trimmed reads from all the samples were searched against the National Center for BiotechnologyInformation (NCBI) nonredundant protein database (version 30.07.2012) (12), using RapSearch2 (32).Details of read statistics for all samples are summarized in Table S1. The data matrix for sample-basedrarefaction curves were computed using the vegan community ecology package (33) by repeatedlysampling the metagenomic reads from sample metagenomes and computing the number of leaves towhich taxa (species level) were assigned (i.e., uncollapsing the taxonomy tree to the species level). Therarefaction curves were then prepared by plotting the taxonomy hits at the species level against thenumber of reads selected randomly. “Leaves” are taxonomic nodes annotated at a given taxonomic level(species in our case).

Metagenome-based taxonomic annotation. We used MEtaGenome ANalyzer (MEGAN) to analyzethe taxonomic profiles of 48 sediment samples collected from two land use types. The resulting outputfiles of paired read sequences were imported and analyzed using the paired-end protocol of MEGAN (34)to obtain taxonomic profiles using the lowest common ancestor (LCA) algorithm. When processing theRapSearch2 output files by using MEGAN, we used parameter settings of “Min Score � 35,” “TopPercent � 10,” “Min Support � 25” and “Minimum sequence complexity threshold � 0.44,” as a thresholdfor species mapping, to create MEGAN-proprietary “rma files” (hereon referred as RMAFs). Reads that didnot have any match to the respective database were placed under the “no hits” node.

16S rRNA-based taxonomic annotation. The 16S rRNA-based taxonomic profiling was conductedusing an in-house-developed pipeline, RiboTagger (35). The metagenomic DNA reads were scanned for33-nucleotide tag sequences from the V6 region as previously described (35, 36). Briefly, a universalprimer (CGACRRCCATGCANCACCT) for the 16S rRNA hypervariable region (V6) was used to scan eachsequencing read to obtain 33-nucleotide sequences downstream of the primer, and the 33-nucleotidesequences were defined as the V6 tag of the 16S rRNA gene. Short reads that did not cover the full33-nucleotide region were discarded. The universal primer matches 94% of 16S rRNA sequences in theRibosomal Database project (RDP) database (37). Each V6 tag was then used as a signature sequence torepresent one operational taxonomic unit.

Functional annotation. Annotated taxonomic assignments were linked to functional roles withKyoto Encyclopedia of Genes and Genomes (KEGG) (38) classifications using MEGAN.

Multiple metagenome comparisons. The output from MEGAN was normalized to the smallest dataset size to allow between-sample comparison of taxonomic abundances for all samples. Comparativeabundance of read counts at different levels of NCBI taxonomy (“class,” “order,” “family,” “genus,” and“species”) and functional categories (KEGG) were exported from all sample comparison files for laterstatistical analyses. We computed unrooted networks from Bray-Curtis distance matrices and a neighbor-net algorithm (39) for different levels of comparative taxonomic and functional profiles as previouslydescribed (40).

Statistical analysis. In order to adjust for differences in sequencing depth among the 48 samples,we subsampled six million reads with replacement from the genus-level annotations from each of the 48data sets. We then computed an unrooted phylogenetic network with SplitsTree4 (41) using theBray-Curtis distance metric. This process was repeated 100 times, generating 100 bootstrappedweighted-edge networks. The total number of nodes in these networks ranged from 707 to 1,273 (meanof 868). To quantify the degree of reproducibility of these networks, we first calculated the shortest pathmatrix for all nodes and then defined the submatrix of the shortest path matrix defined by the 48 samplenodes. Within- and between-group summary distances were subsequently computed using the methodsof Gower and Krzanowski (42). All computations were performed using R version 3.0.2 (43).

Taxonomic and functional profiles (normalized read counts of taxa or orthologous genes exportedfrom MEGAN) of the microbial community were compared between both land use types (industrialversus residential) and rain perturbation (“pre” versus “post” rain) using the permutational multivariateanalysis of variance (PERMANOVA) (44) add-on in Plymouth Routines In Multivariate Ecological Research(PRIMER-v6) (45). For such analyses, we considered the taxonomic profile of the full data set at the genuslevel and functional profiles (KEGG levels 2 and 3). Each data set was square root transformed prior tocalculating a Bray-Curtis similarity matrix. Similarity percentage analysis (SIMPER) analyses (46) were usedto determine which taxon/function contributed most to the observed differences. In addition, dispersion(variability) of the multivariate data was compared between land use types and before versus after rain

Saxena et al.

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 12

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

using permutational multivariate analyses of DISPersion (PERMDISP) (45). Finally, “relate analysis” wasused to determine relationships between taxon and function similarity matrices (47). The significantlydifferent taxonomic groups, functional genes, and differences in the diversity indices are calculatedbased on t test with standard Bonferroni correction and 999 permutations. The significance of thedifferences between the diversities of microbial communities from urban waterways and microbialcommunities from a previously published study of Nanxijiang River near Wenzhou city in the ZhejiangProvince of China (13) were computed using Mann-Whitney U test.

Data availability. The metagenome sequencing data sets supporting the conclusions of this articleare available in the NCBI Sequence Read Archive (SRA) study under accession number SRP051069 andStudy BioProject number PRJNA267173.

SUPPLEMENTAL MATERIALSupplemental material for this article may be found at https://doi.org/10.1128/

mSystems.00136-17.TEXT S1, DOCX file, 0.01 MB.FIG S1, TIF file, 2.8 MB.FIG S2, TIF file, 2.9 MB.FIG S3, TIF file, 2.8 MB.TABLE S1, DOCX file, 0.01 MB.TABLE S2, DOCX file, 0.01 MB.TABLE S3, DOCX file, 0.01 MB.TABLE S4, DOCX file, 0.01 MB.TABLE S5, DOCX file, 0.01 MB.

ACKNOWLEDGMENTSWe thank Daniela Moses and Stephan C. Schuster for assistance with sequencing

and insightful discussions and Krithika Arumugam for bioinformatic support.

REFERENCES1. NASA. 2014. NASA Earth Observatory. EOS Project Science Office, NASA

Goddard Space Flight Center, Greenbelt, MD. http://earthobservatory.nasa.gov/GlobalMaps/view.php?d1�TRMM_3B43M.

2. United Nations. 2013. World urbanization prospects the 2012 revision.Population Division, Department of Economic and Social Affairs, UnitedNations, New York, NY.

3. Wong TH. 2006. An overview of water sensitive urban design practicesin Australia. Water Pract Technol 1:1– 8.

4. Tortajada C, Joshi YK. 2013. Water demand management in Singapore:involving the public. Water Resour Manag 27:2729–2746. https://doi.org/10.1007/s11269-013-0312-5.

5. Tortajada C. 2006. Water management in Singapore. Int J Water ResourDev 22:227–240. https://doi.org/10.1080/07900620600691944.

6. Petersen DG, Blazewicz SJ, Firestone M, Herman DJ, Turetsky M, WaldropM. 2012. Abundance of microbial genes associated with nitrogen cyclingas indices of biogeochemical process rates across a vegetation gradientin Alaska. Environ Microbiol 14:993–1008. https://doi.org/10.1111/j.1462-2920.2011.02679.x.

7. Reese BK, Mills HJ, Dowd SE, Morse JW. 2013. Linking molecular micro-bial ecology to geochemistry in a coastal hypoxic zone. Geomicrobiol J30:160 –172. https://doi.org/10.1080/01490451.2012.659331.

8. El-Sheekh MM, Hamouda RA. 2013. Biodegradation of crude oil by somecyanobacteria under heterotrophic conditions. Desalin Water Treat 52:7–9, 1448 –1454. https://doi.org/10.1080/19443994.2013.794008.

9. Saxena G, Marzinelli EM, Naing NN, He Z, Liang Y, Tom L, Mitra S, PingH, Joshi UM, Reuben S, Mynampati KC, Mishra S, Umashankar S, Zhou J,Andersen GL, Kjelleberg S, Swarup S. 2015. Ecogenomics reveals metalsand land-use pressures on microbial communities in the waterways ofa megacity. Environ Sci Technol 49:1462–1471. https://doi.org/10.1021/es504531s.

10. Badin AL, Faure P, Bedell JP, Delolme C. 2008. Distribution of organicpollutants and natural organic matter in urban storm water sediments asa function of grain size. Sci Total Environ 403:178 –187. https://doi.org/10.1016/j.scitotenv.2008.05.022.

11. Walsh CJ. 2000. Urban impacts on the ecology of receiving waters: aframework for assessment, conservation and restoration. Hydrobiologia431:107–114. https://doi.org/10.1023/A:1004029715627.

12. Benson DA, Karsch-Mizrachi I, Lipman DJ, Ostell J, Wheeler DL. 2005.GenBank. Nucleic Acids Res 33:D34 –D38. https://doi.org/10.1093/nar/gki063.

13. Lu X-M, Lu P-Z. 2014. Characterization of bacterial communities insediments receiving various wastewater effluents with high-throughputsequencing analysis. Microb Ecol 67:612– 623. https://doi.org/10.1007/s00248-014-0370-0.

14. Peng J, Li H, Su J, Zhang Q, Rui J, Cai C. 2013. Response of bacterialcommunities to short-term pyrene exposure in red soil. Front Environ SciEng 7:559 –567. https://doi.org/10.1007/s11783-013-0501-8.

15. Ledin M. 2000. Accumulation of metals by microorganisms — processesand importance for soil systems. Earth-Sci Rev 51:1–31. https://doi.org/10.1016/S0012-8252(00)00008-8.

16. Kirchman DL, Dittel AI, Findlay SE, Fischer D. 2004. Changes in bacterialactivity and community structure in response to dissolved organic matter inthe Hudson River, New York. Aquat Microb Ecol 35:243–257. https://doi.org/10.3354/ame035243.

17. Rochex A, Godon J-J, Bernet N, Escudié R. 2008. Role of shear stress oncomposition, diversity and dynamics of biofilm bacterial communities.Water Res 42:4915– 4922. https://doi.org/10.1016/j.watres.2008.09.015.

18. Liu Y, Tay J-H. 2002. The essential role of hydrodynamic shear force inthe formation of biofilm and granular sludge. Water Res 36:1653–1665.https://doi.org/10.1016/S0043-1354(01)00379-7.

19. Cao YS, Alaerts GJ. 1995. Influence of reactor type and shear stress onaerobic biofilm morphology, population and kinetics. Water Res 29:107–118. https://doi.org/10.1016/0043-1354(94)00136-U.

20. van der Grinten E, Pikkemaat MG, van den Brandhof E-J, Stroomberg GJ,Kraak MHS. 2010. Comparing the sensitivity of algal, cyanobacterial andbacterial bioassays to different groups of antibiotics. Chemosphere 80:1– 6. https://doi.org/10.1016/j.chemosphere.2010.04.011.

21. DiDonato RJ, Young ND, Butler JE, Chin K-J, Hixson KK, Mouser P, LiptonMS, DeBoy R, Methé BA. 2010. Genome sequence of the deltaproteo-bacterial strain NaphS2 and analysis of differential gene expressionduring anaerobic growth on naphthalene. PLoS One 5:e14072. https://doi.org/10.1371/journal.pone.0014072.

22. Strycharz SM, Gannon SM, Boles AR, Franks AE, Nevin KP, Lovley DR.2010. Reductive dechlorination of 2-chlorophenol by Anaeromyxo-

Urban Waterway Microbial Communities

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 13

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from

bacter dehalogenans with an electrode serving as the electron donor.Environ Microbiol Rep 2:289 –294. https://doi.org/10.1111/j.1758-2229.2009.00118.x.

23. Daims H, Lebedeva EV, Pjevac P, Han P, Herbold C, Albertsen M, JehmlichN, Palatinszky M, Vierheilig J, Bulaev A, Kirkegaard RH, von Bergen M,Rattei T, Bendinger B, Nielsen PH, Wagner M. 2015. Complete nitrifica-tion by Nitrospira bacteria. Nature 528:504 –509. https://doi.org/10.1038/nature16461.

24. Steeneveld GJ, Koopmans S, Heusinkveld BG, Theeuwes NE. 2014. Re-freshing the role of open water surfaces on mitigating the maximumurban heat island effect. Landscape Urban Planning 121:92–96. https://doi.org/10.1016/j.landurbplan.2013.09.001.

25. Liang Y, Van Nostrand JD, Deng Y, He Z, Wu L, Zhang X, Li G, Zhou J.2011. Functional gene diversity of soil microbial communities from fiveoil-contaminated fields in China. ISME J 5:403– 413. https://doi.org/10.1038/ismej.2010.142.

26. Winter C, Hein T, Kavka G, Mach RL, Farnleitner AH. 2007. Longitudinalchanges in the bacterial community composition of the Danube River: awhole-river approach. Appl Environ Microbiol 73:421– 431. https://doi.org/10.1128/AEM.01849-06.

27. Menon R, Jackson CR, Holland MM. 2013. The influence of vegetation onmicrobial enzyme activity and bacterial community structure in fresh-water constructed wetland sediments. Wetlands 33:365–378. https://doi.org/10.1007/s13157-013-0394-0.

28. Tamber S, Ochs MM, Hancock REW. 2006. Role of the novel OprD familyof porins in nutrient uptake in Pseudomonas aeruginosa. J Bacteriol188:45–54. https://doi.org/10.1128/JB.188.1.45-54.2006.

29. Olaitan AO, Morand S, Rolain JM. 2014. Mechanisms of polymyxinresistance: acquired and intrinsic resistance in bacteria. Front Microbiol5:643. https://doi.org/10.3389/fmicb.2014.00643.

30. Nikaido H. 2009. Multidrug resistance in bacteria. Annu Rev Biochem 78:119–146. https://doi.org/10.1146/annurev.biochem.78.082907.145923.

31. Martin M. 2011. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet.journal 17:10 –12. https://doi.org/10.14806/ej.17.1.200.

32. Zhao Y, Tang H, Ye Y. 2012. RAPSearch2: a fast and memory-efficientprotein similarity search tool for next-generation sequencing data. Bioin-formatics 28:125–126. https://doi.org/10.1093/bioinformatics/btr595.

33. Oksanen J, Blanchet FG, Kindt R, Legendre P, Minchin PR, O’hara RB,Simpson GL, Solymos P, Stevens MHH, Wagner H. 2013. Package ‘vegan.Community Ecology Package, version 2.

34. Huson DH, Mitra S, Ruscheweyh H-J, Weber N, Schuster SC. 2011.

Integrative analysis of environmental sequences using MEGAN4. Ge-nome Res 21:1552–1560. https://doi.org/10.1101/gr.120618.111.

35. Tan CH, Koh KS, Xie C, Tay M, Zhou Y, Williams R, Ng WJ, Rice SA,Kjelleberg S. 2014. The role of quorum sensing signalling in EPS pro-duction and the assembly of a sludge community into aerobic granules.ISME J 8:1186 –1197. https://doi.org/10.1038/ismej.2013.240.

36. Xie C, Goi CLW, Huson DH, Little PFR, Williams RBH. 2016. RiboTagger:fast and unbiased 16S/18S profiling using whole community shotgunmetagenomic or metatranscriptome surveys. BMC Bioinformatics 17:508. https://doi.org/10.1186/s12859-016-1378-x.

37. Claesson MJ, Wang Q, O’Sullivan O, Greene-Diniz R, Cole JR, Ross RP,O’Toole PW. 2010. Comparison of two next-generation sequencing tech-nologies for resolving highly complex microbiota composition usingtandem variable 16S rRNA gene regions. Nucleic Acids Res 38:e200.https://doi.org/10.1093/nar/gkq873.

38. Kanehisa M, Goto S. 2000. KEGG: Kyoto Encyclopedia of Genes and Ge-nomes. Nucleic Acids Res 28:27–30. https://doi.org/10.1093/nar/28.1.27.

39. Bryant D, Moulton V. 2004. Neighbor-Net: an agglomerative method forthe construction of phylogenetic networks. Mol Biol Evol 21:255–265.https://doi.org/10.1093/molbev/msh018.

40. Mitra S, Gilbert JA, Field D, Huson DH. 2010. Comparison of multiplemetagenomes using phylogenetic networks based on ecological indices.ISME J 4:1236 –1242. https://doi.org/10.1038/ismej.2010.51.

41. Huson DH, Bryant D. 2006. Application of phylogenetic networks inevolutionary studies. Mol Biol Evol 23:254 –267. https://doi.org/10.1093/molbev/msj030.

42. Gower JC, Krzanowski WJ. 1999. Analysis of distance for structuredmultivariate data and extensions to multivariate analysis of variance. J RStatist Soc C 48:505–519. https://doi.org/10.1111/1467-9876.00168.

43. R Core Team. 2013. R: a language and environment for statistical com-puting. R Foundation for Statistical Computing, Vienna, Austria.

44. Anderson MJ. 2001. A new method for non-parametric multivariate anal-ysis of variance. Austral Ecol 26:32–46. https://doi.org/10.1046/j.1442-9993.2001.01070.x.

45. Anderson MJ, Gorley RN, Clarke RK. 2008. Permanova� for primer: guideto software and statistical methods. Primer-E Ltd, Plymouth, UnitedKingdom.

46. Clarke KR. 1993. Non-parametric multivariate analyses of changes incommunity structure. Austral Ecol 18:117–143. https://doi.org/10.1111/j.1442-9993.1993.tb00438.x.

47. Clarke KR, Clarke RK, Gorley RN. 2006. Primer V6: user manual—tutorial.Plymouth Marine Laboratory, Plymouth, United Kingdom.

Saxena et al.

May/June 2018 Volume 3 Issue 3 e00136-17 msystems.asm.org 14

on June 19, 2020 by guesthttp://m

systems.asm

.org/D

ownloaded from